AMADEUS: an adaptive multi-agent system to learn a user’s recurring actions in ambient systems

Ambient systems are characterized by their dynamics and their huge complexity. An important issue in this field is their capability to provide a relevant behaviour in order to satisfy users involved. Multi-agent systems, because of their ability to deal with dynamic, distributed and not deterministic environments, seem to be very promising to solve adaptation problems in ambient systems. The objective of our study is to propose Amadeus, a system able to learn the user’s behaviour in order to perform his recurrent actions on his behalf, independently of the ambient system in which it is applied. The originality of our contribution is to be generic and to promote a process able to learn at runtime without any prior learning phase and able to filter useful data for characterizing users' context.